Harnessing Deep Learning for TomoSAR stack enhancement

Inhalt:
Synthetic aperture radar (SAR) tomography (TomoSAR) utilizes co-registered SAR images from different tracks (known as a “TomoSAR stack”) to create a resolution in the height direction. Spectral analysis retrieves the vertical backscattered Power Spectrum Pattern, enabling 3D imaging (Tomography). Tomograms show ambiguities inversely related to baseline separation, with larger and denser stacks offering better ambiguity rejection. To address the constraints posed by a limited number of acquisitions, we utilize a deep neural network. This network is employed to synthesize Single Look Complex SAR images by introducing an "artificial" baseline that was not part of the original TomoSAR stack.